On the arithmetic size of linear differential equations (Q5945611): Difference between revisions

This paper generalizes the notion of \(G\) function from the context of formal power series over a number field which satisfy a differential equation and have a positive radius of convergence for all complex embeddings to connections. Let \(K\) be a number field, \(V_K\) its ring of integers and \(S=\text{Spec}(V_S) \subset \text{Spec}(V_K)\) an open subscheme. Let \(\Sigma_S\) denote the set of closed points of \(S\). Let \(F\) be a function field over \(K\); a smooth \(S\) model of \(F\) is a smooth \(S\) scheme \(f:X \to S\) of finite type with geometrically connected non-empty fibers, and such that the function field of the generic fibre is \(F\). Let \((M,\nabla)\) be an \(F/K\) differential module of finite \(F\) rank with connection \(\nabla: M \to \Omega^1_{F/K} \otimes M\). A model of \((M,\nabla)\) on \(X/S\) is a locally free \(\mathcal O_X\) module \(\mathcal M\) with an integrable connection \(\nabla:\mathcal M \to \Omega^1_{X/S} \otimes \mathcal M\) whose stalk at the generic point of \(X\) is \((M,\nabla)\). For a closed point \(\nu\) of \(S\), let \(p(\nu)\) denote the characteristic of the residue field \(k(\nu)\) and let \(\eta_\nu\) denote the generic point of the closed fibre of \(f\) over it. There is a natural way to extend the valuation corresponding to \(\nu\) to \(F\) such that the stalk of \(\mathcal O_X\) at \(\eta_\nu\) is the set of elements of value at most one. Using a model of \((M,\nabla)\), the author defines a generic \(\nu\)-adic radius of convergence of \((M,\nabla)\), and calls the sum of the logs of their reciprocals the global inverse radius. The module is said to be of type \(G\) if this latter sum is finite. The author further defines another invariant, called the size, associated to \((M,\nabla)\). The main theorem then establishes upper and lower bounds for the difference between generic radius of convergence and size (in the case the former is finite) in terms of an expression involving the \(p(\nu)\)'s for \(\nu\) such that the induced connection on the fibre over \(\nu\) has a certain property (is not \(p(\nu)\) nilpotent of exponent \(\leq 1\)). Among other things, this establishes that bounded size is equivalent to bounded global inverse radius, and that the difference of the two is independent of the choice of model \((\mathcal M, \nabla)\), and the upper and lower bound for the difference is independent of the choice of \(S\).